See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/269211816

Computational fluid dynamics for aerodynamic design - Its current and future
impact

Conference Paper · January 2001

DOI: 10.2514/6.2001-538

CITATIONS READS
76 247

2 authors:

Antony Jameson J. Vassberg

Stanford University The Boeing Company Retired
474 PUBLICATIONS 28,779 CITATIONS 95 PUBLICATIONS 3,565 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Towards parameter-free subgrid-scale modelling in large eddy simulation View project

Multigrid smoothers for compressible turbulent flow simulations View project

All content following this page was uploaded by J. Vassberg on 06 May 2020.

The user has requested enhancement of the downloaded file.

 AIAA-2001-0538
Computational Fluid Dynamics
for Aerodynamic Design:
Its Current and Future Impact

A. Jameson
Department of Aeronautics and Astronautics
Stanford University
Stanford, CA 94305, USA
J. C. Vassberg
Phantom Works
The Boeing Company
Long Beach, CA 90807, USA

39th AIAA Aerospace Sciences Meeting & Exhibit

8-11 January, 2001 / Reno, NV

For permission to copy or republish, contact the American Institute of Aeronautics and Astronautics
1801 Alexander Bell Drive, Suite 500, Reston, Va. 22091
 Computational Fluid Dynamics for Aerodynamic Design:
Its Current and Future Impact
Antony Jameson  John C. Vassberg y
Department of Aeronautics and Astronautics Phantom Works
Stanford University The Boeing Company
Stanford, CA 94305, USA Long Beach, CA 90807, USA

8 11 January, 2001

Abstract The paper nishes with some visions for the fu-
ture. Extrapolating the trends of computer weight
This paper discusses the role that computational and cost, it is interesting to speculate on how the
uid dynamics plays in the design of aircraft. An aircraft design environment may evolve in the years
overview of the design process is provided, covering to come.
some of the typical decisions that a design team ad-
dresses within a multi-disciplinary environment. On
a very regular basis trade-o s between disciplines Nomenclature
have to be made where a set of con icting require-
ments exist. Within an aircraft development project, A Hessian Matrix / Operator
we focus on the aerodynamic design problem and re- AR Wing Aspect Ratio = Sbref2

view how this process has been advanced, rst with

the improving capabilities of traditional computa- b Wing Span
tional uid dynamics analyses, and then with aero- B Shape Function Basis
dynamic optimizations based on these increasingly CFD Computational Fluid Dynamics
accurate methods. The optimization method of the CD Drag Coecient = q1DragSref
present work is based on the use of the adjoint of the CL Lift Coecient = q1 Sref
Lift
ow equations to compute the gradient of the cost
function. Then, we use this gradient to navigate the Cref Wing Reference Chord
design space in an ecient manner to nd a local count Drag Coecient Unit = 0:0001
minimum. The computational costs of the present F Surface De ning Function
method are compared with that of other approaches G Acceleration Unit = Gravity ' 32:2 sec
ft
2
to aerodynamic optimization. A brief discussion re- G Gradient of Cost Function
garding the formulation of a continuous adjoint, as H Estimate of Inverse Hessian Matrix
opposed to a discrete one, is also included.
Two case studies are provided which highlight the I Objective or Cost Function
bene ts of utilizing automatic aerodynamic shape N Number of Design Variables
optimization. The rst case chronicles the appli- Re Wing Reynolds number based on Cref
cation of such software during a complete aircraft Re Attachment Line Reynolds number
design e ort, while the second case shows to what Sref Wing Reference Area
extent the aerodynamic design cycle can be com- x Independent Spatial Variable
pressed. In both e orts, the optimizations were per- q Dynamic Pressure = 12 V 2
formed in design spaces with dimensions greater than
4; 000. Furthermore, they were conducted on very  Wing Taper Ratio
a ordable computer equipment and turned around c=4 Wing Quarter-Chord Sweep
within a few hours of wall-clock time.  3:141592654:::
 AIAA Fellow, T. V. Jones Prof. of Engineering 1 In nity
y AIAA Associate Fellow, Boeing Technical Fellow  First Variation of
O() Order of
Copyright c 2001 by Jameson & Vassberg.
Published by the AIAA with permission. () 1 Inverse Matrix of

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 1 of 26

1 Introduction
The past 25 years have seen a revolution in the entire
engineering design process as computational simula-
tion has come to play an increasingly dominant role.
Today, engineers spend most of their time at work-
stations.
Most notably, computer aided design (CAD)
methods have essentially replaced the drawing board
as the basic tool for de nition and control of the
con guration. Software systems such as CATIA and
Unigraphics provide solid modeling capabilities that
enable designers to prepare complex layouts without
the need to build mockups. Computer visualization
techniques enable the designer to verify that no in-
terferences exist between di erent parts in the lay-
out, and greatly facilitate decisions on the routing of
electrical wiring and hydraulic piping.
Similarly, structural analysis is now almost en-
tirely carried out by computational methods typi-
cally based on the nite element method. Com-
mercially available software systems such as NAS-
TRAN, ANSYS, or ELFINI have been progressively
developed and augmented with new features, and
can treat the full range of requirements for aeronau-
tical structures, including analysis of stressed skin
into the nonlinear range. Also, they are very care-
fully validated against a comprehensive suite of test
cases before each new release. Hence, engineers place
complete con dence in their results. Accordingly, the
structural design is routinely committed on the basis
of computational analysis, while structural testing is
limited to the role of veri cation that the design truly
meets its speci ed requirements of ultimate strength
and fatigue life.
Computational simulation of uid ow has not yet
reached the same level of maturity. While commer-
cial software for the simulation of uid ow is of-
fered by numerous vendors, aircraft companies con-
tinue to make substantial investments in the in-house
development of their own methods, such as Boeing's
TRANAIR, or Lockheed's SPLITFLOW and TEAM
programs. At the same time there are major ongoing
e orts to develop the science of computational uid
dynamics (CFD) in government research agencies
such as NASA, Japan's ARL, or in Europe, France's
ONERA, Germany's DLR, Holland's NLR and Swe-
den's FFA, all of which are a source of industrially
used computer programs. This re ects the fact that
uid ow is generally more complex and harder to
predict than the behavior of structures. The com-
plexity and range of phenomena of uid ow is well
illustrated in Van Dyke's Album of Fluid Motion [1].
All this e ort has led to major advances. De-
spite these, CFD is still not being exploited as ef-
fectively as one would like in the design process.
Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV
This is partially due to the long set-up times and
high costs, both human and computational, asso-
ciated with complex ow simulations. This paper
examines ways to exploit computational simulation
more e ectively in the overall design process, with
the primary focus on aerodynamic design, while rec-
ognizing that this should be part of an integrated
multi-disciplinary process. The design process itself
is surveyed in the next section. The following section
examines the way in which optimization techniques
can be integrated with CFD. The paper concludes
with examples of two case studies which apply aero-
dynamic shape optimization methods within a design
environment.
2 Design Process
The design process can generally be divided into
three phases: conceptual design, preliminary design,
and nal detailed design, as illustrated in Figure 1.
The conceptual design stage de nes the mission in
the light of anticipated market requirements, and de-
termines a general preliminary con guration capable
of performing this mission, together with rst esti-
mates of size, weight and performance. In the prelim-
inary design stage the aerodynamic shape and struc-
tural skeleton progress to the point where detailed
performance estimates can be made and guaranteed
to potential customers, which can then, in turn, for-
mally sign binding contracts for the purchase of a
certain number of aircraft. At this stage the devel-
opment costs are still fairly moderate, in the range
of 50 - 100 million dollars. In the nal design stage
the structure must be de ned in complete detail, to-
gether with complete systems, including the ight
deck, control systems (involving major software de-
velopment for y-by-wire systems), electrical and hy-
draulic systems, landing gear, weapon systems for
military aircraft, and cabin layout for commercial
aircraft. Major costs are incurred at this stage, dur-
ing which it is also necessary to prepare a detailed
manufacturing plan, together with appropriate facil-
ities and tooling. The development costs to reach the
point of initial production are in the range of 3 - 10
billion dollars. Thus, the nal design would normally
be carried out only if sucient orders have been re-
ceived to indicate a reasonably high probability of
recovering a signi cant fraction of the investment.
In the development of commercial aircraft, aerody-
namic design plays a leading role in the preliminary
design stage. The de nition of the external aerody-
namic shape may actually be nalized in the pre-
liminary design. The aerodynamic lines of the Boe-
ing 777 were frozen, for example, when initial orders
were accepted before the initiation of the detailed
2 of 26
 Conceptual Defines Mission Conceptual
Preliminary sizing Design
Design Weight, performance

CAD Central Database

Definition

Mesh
Generation Detailed
Preliminary Final

Major Design Cycle

Design Design

Outer Loop
CFD

Inner Loop
Analysis

Visualization
Release to
Manufacturing

Final Design Performance

Evaluation

{
Propulsion
Noise
Stability
Multi−Disciplinary Control
Evaluation Loads
Structures
Fabrication

Figure 1: The Overall Design Process Wind Tunnel

Testing
Model
Fabrication

design of the structure. Figure 2 illustrates the way

in which the aerodynamic design process is embed-
ded in the overall preliminary design. The starting
point is an initial CAD de nition resulting from the
conceptual design. The inner loop of aerodynamic
analysis is contained in an outer multi-disciplinary
loop, which is in turn contained in a major design cy-
cle involving wind tunnel testing. In recent Boeing
practice, three major design cycles, each requiring
about 4-6 months, have been used to nalize the wing
design. Improvements in CFD which would allow
the elimination of a major cycle would signi cantly
shorten the overall design process and therefore re-
duce costs.
The inner aerodynamic design loop is used to eval-
uate numerous variations in the wing de nition. In
each iteration it is necessary to generate a mesh for
the new con guration prior to performing the CFD
analysis. Computer graphics software is then used
to visualize the results, and the performance is eval-
uated. The rst studies may be con ned to par-
tial con gurations such as wing-body or wing-body-
nacelle combinations. At this stage the focus is on
the design of the clean wing. Key points of the ight
envelope include the nominal cruise point, cruise at
high lift and low lift to allow for the weight varia-
tion between the initial and nal cruise as the fuel
is burned o , and a long range cruise point at lower
Mach number, where it is important to make sure
there is no signi cant drag creep. Other de ning
points are the climb condition, which requires a good
lift to drag ratio at low Mach number and high lift
coecient with a clean wing, and the bu et condi-
tion. The bu et requirement is typically taken as
Figure 2: The Aerodynamic Design Process
the high lift cruise point increased to a load of 1.3 g
to allow for maneuvering and gust loads. Both wing
section modi cations such as the thickness to chord
ratio and camber distributions, and planform vari-
ations such as the sweep-back angle or aspect ratio
may be considered. While the detailed design of the
high lift system and control surfaces may be deferred
to a later stage, the planform must provide the nec-
essary space for both high lift systems and control
surfaces outside the main structural box, and it must
also accommodate the landing gear. This generally
requires an extension of the inboard trailing edge to
form an area known as a "yehudi".
The aerodynamic analysis interacts with the other
disciplines in the next outer loop. These disciplines
have their own inner loops, not shown in Figure 2.
For an ecient design process the fully updated aero-
design database must be accessible to other disci-
plines without loss of information. For example, the
thrust requirements in the power plant design will
depend on the drag estimates for take-o , climb and
cruise. In order to meet airport noise constraints a
rapid climb may be required while the thrust may
also be limited. Initial estimates of the lift and mo-
ments allow preliminary sizing of the horizontal and
vertical tail. This interacts with the design of the
control system, where the use of a y-by-wire system
may allow relaxed static stability and tail surfaces of
reduced size.

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 3 of 26

 In fact, the interaction between disciplines, as well
as the e ect of disciplinary constraints on the other
participating disciplines is quite strong. This inter-
action between disciplines should not be reserved to
the analysis of the coupled system, but should be
extended to the computation of coupled sensitivities
to be used during the design process. For exam-
ple, the aerodynamic analysis may allow for shape
changes due to aeroelastic e ects. As this multidis-
ciplinary process becomes more closely integrated,
the second loop in Figure 2 will slowly disappear.
Aero-structural design is only an example of the in-
teraction between disciplines. Additional disciplines
may have as strong an impact on the design of the
coupled system and will have to be considered in a
way which will be much di erent from sequential in-
teraction. In fact, this may be also true in the design
of systems di erent from aircraft (spacecraft, auto-
mobiles, ships, chip manufacturing and layout, etc.).
With this in mind, rst estimates of the aerody-
namic loads allow the design of an initial structural
skeleton, which in turn provides an estimate of the
structure weight. One of the main trade-o s is be-
tween aerodynamic performance and wing structure
weight. The requirement for fuel volume may also
be an important consideration. An increase in the
thickness to chord ratio both increases fuel volume,
and allows the same bending moment to be carried
with reduced skin thickness, with an accompanying
reduction in weight. On the other hand it will lead
to a decrease in the drag rise Mach number. The
induced drag, which typically contributes around 40
percent of the cruising drag, varies inversely as the
square of the span. Thus a 5 percent increase in
the wing span could produce a total drag reduction
of the order of 4 percent, but would lead to an in-
crease in wing weight because of the increase in the
root bending moment. The wing span may in fact
be limited by airport gate constraints.
The taper ratio and span load distribution also af-
fect the trade-o between aerodynamic performance
and wing weight. While an elliptic span load distri-
bution minimizes the induced drag for a given span,
a more triangular load distribution reduces the root
bending moment. A large root chord may be dic-
tated by the need to accommodate the landing gear
and aps, but it also has the advantage of increasing
the root thickness for a xed thickness to chord ra-
tio, yielding a weight reduction. In order to maintain
a moderately ecient span load distribution with a
highly tapered planform the outboard wing must op-
erate with higher local section lift coecient than
the inboard wing. This can have an adverse e ect
on the behavior of a swept-back wing near bu et, as
the outboard wing may incur a shock-induced stall
before the inboard wing, leading to a reduction of
Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV
lift behind the center of gravity, and consequently a
high speed pitch-up. This will be unacceptable for
certi cation if it is too severe.
An increase in the wing sweep-back angle may be
used to increase the drag rise Mach number. Alterna-
tively it allows an increase in the thickness to chord
ratio for the same drag rise Mach number, with a re-
sulting weight reduction. This is partially o set by
the increase in the length of the wing. Moreover, an
increase in the sweep back angle will aggravate the
problem of high speed pitch-up. Most modern highly
loaded wings have sweep back angles no greater than
35 degrees at the 14 chord line.
Manufacturing constraints must also be consid-
ered in the nal de nition of the aerodynamic shape.
For example, the curvature in the spanwise direc-
tion should be limited. This avoids the need for shot
peaning which might otherwise be required to pro-
duce curvature in both the spanwise and chordwise
directions.
In order to carry out the inner loop of the aero-
dynamic design process the main requirements for
e ective CFD software are:
1. Sucient and known level of accuracy
2. Acceptable computational and manpower costs
3. Fast turn around time
Performance estimation at the cruise condition is
crucial to the design of long-range transport aircraft.
The error should be on the order of 1 percent. The
drag coecient of a typical transport aircraft such as
the Boeing 747 is about 0.0275 (275 counts), depend-
ing on the lift coecient, which is approximately 0.5.
The drag coecient of proposed supersonic transport
designs is in the range of 0.0120 to 0.0150 at much
lower lift coecients in the range of 0.1 to 0.12. Thus
one should aim to predict drag with an accuracy of 1
to 2 counts. Manufacturers have to guarantee perfor-
mance, and errors can be very expensive through the
costs of redesign, penalty payments and lost orders.
In order to achieve this level of accuracy, it is
ultimately essential to use the Reynolds-Averaged
Navier-Stokes (RANS) equations. However, to ac-
celerate their initial e orts, designers typically use
CFD methods based on less sophisticated ow mod-
els, such as the full-potential or Euler equations cou-
pled with a boundary-layer method. In order to allow
the completion of the major design cycle in 4 - 6
months, the cycle time for the multidisciplinary loop
should not be greater than about 2 weeks. Consid-
ering the need to examine the performance of design
variations at all the key points of the ight envelope,
this implies the need to turn around aerodynamic
analyses in a few hours. The computational costs
4 of 26
are also important because the cumulative costs of
large numbers of calculations can become a limiting
factor.
It is also evident that the number of possible design
variations is too large to permit their exhaustive eval-
uation, and thus it is very unlikely that a truly op-
timum solution can be found without the assistance
of automatic optimization procedures. Ultimately
there is a need for multi-disciplinary optimization
(MDO), but this can only be e ective if it is based
on suciently high delity modeling of the separate
disciplines. As a step in this direction there could
be signi cant pay-o s from the application of opti-
mization techniques within the disciplines, where the
interactions with other disciplines are taken into ac-
count through the introduction of constraints. For
example the wing drag can be minimized at a given
Mach number and lift coecient with a xed plan-
form, and constraints on minimum thickness to meet
requirements for fuel volume and structure weight.
3 Aerodynamic Optimization
Traditionally the process of selecting design varia-
tions has been carried out by trial and error, relying
on the intuition and experience of the designer. It is
not at all likely that repeated trials in an interactive
design and analysis procedure can lead to a truly op-
timum design. In order to take full advantage of the
possibility of examining a large design space the nu-
merical simulations need to be combined with auto-
matic search and optimization procedures. This can
lead to automatic design methods which will fully
realize the potential improvements in aerodynamic
eciency.
An approach which has become increasingly pop-
ular is to carry out a search over a large number
of variations via a genetic algorithm. This may
allow the discovery of (sometimes unexpected) opti-
mum design choices in very complex multi-objective
problems, but it becomes extremely expensive when
each evaluation of the cost function requires intensive
computation, as is the case in aerodynamic problems.
In order to nd optimum aerodynamic shapes with
reasonable computational costs, it pays to embed the
ow physics within the optimization process. In fact,
one may regard a wing as a device to control the ow
in order to produce lift with minimum drag. As a
result, one can draw on concepts which have been
developed in the mathematical theory of control of
systems governed by partial di erential equations. In
particular, an acceptable aerodynamic design must
have characteristics that smoothly vary with small
changes in shape and ow conditions. Consequently,
gradient-based procedures are appropriate for aero-
Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV
dynamic shape optimization. Two main issues a ect
the eciency of gradient-based procedures; the rst
is the actual calculation of the gradient, and the sec-
ond is the construction of an ecient search proce-
dure which utilizes the gradient.
3.1 Gradient Calculation
For the class of aerodynamic optimization problems
under consideration, the design space is essentially
in nitely dimensional. Suppose that the perfor-
mance of a system design can be measured by a cost
function I which depends on a function F (x) that
describes the shape, where under a variation of the
design,  F (x), the variation of the cost is I . Now
suppose that I can be expressed to rst order as
Z
I = G (x) F (x)dx
where G (x) is the gradient. Then by setting
 F (x) = G (x)
one obtains an improvement
Z
I =  G 2 (x)dx
unless G (x) = 0: Thus the vanishing of the gradient
is a necessary condition for a local minimum.
Computing the gradient of a cost function for a
complex system can be a numerically intensive task,
especially if the number of design parameters is large
and if the cost function is an expensive evaluation.
The simplest approach to optimization is to de ne
the geometry through a set of design parameters,
which may, for example, be the weights i applied
to a set of shape functions Bi (x) so that the shape is
represented as
X
F (x) = iBi (x):
Then a cost function I is selected which might be
the drag coecient or the lift to drag ratio; I is
regarded as a function of the parameters i. The
sensitivities @@Ii may now be estimated by making a
small variation  i in each design parameter in turn
and recalculating the ow to obtain the change in I .
Then
@I I ( i +  i ) I ( i ) :
@ i
 i
The main disadvantage of this nite-di erence ap-
proach is that the number of ow calculations needed
to estimate the gradient is proportional to the num-
ber of design variables [2]. Similarly, if one resorts
to direct code di erentiation (ADIFOR [3, 4]), or
5 of 26
complex-variable perturbations [5], the cost of de- where T  
termining the gradient is also directly proportional G = @I
@F
T @R :
@F
to the number of variables used to de ne the design.
Even small problems of aerodynamic shape optimiza- The advantage is that the variation in cost function is
tion based on these approaches can require compute independent of w, with the result that the gradient
resources that are measured in CPU-Years, which of I with respect to any number of design variables
can only be completed in reasonable elapsed time can be determined without the need for additional
through utilization of massively-parallel computers ow- eld evaluations.
costing millions of dollars. In the case that (1) is a partial di erential equa-
A more cost e ective technique is to compute the tion, the adjoint equation (2) is also a partial di er-
gradient through the solution of an adjoint problem, ential equation and appropriate boundary conditions
such as that developed by the authors [6, 7, 8]. The must be determined. It turns out that the appro-
essential idea may be summarized as follows. For priate boundary conditions depend on the choice of
ow about an arbitrary body, the aerodynamic prop- the cost function, and may easily be derived for cost
erties that de ne the cost function are functions of functions that involve surface-pressure integrations.
the ow eld variables (w) and the physical shape of Cost functions involving eld integrals lead to the
the body, which may be represented by the function appearance of a source term in the adjoint equation.
F . Then The cost of solving the adjoint equation is com-
I = I (w; F ) parable to that of solving the ow equation. Hence,
and a change in F results in a change of the cost the cost of obtaining the gradient is comparable to
function the cost of two function evaluations, regardless of the
@I T @I T dimension of the design space. The downside of this
I = @w w + @ F  F : approach is that it can take man-months to develop
Using a technique drawn from control theory, the an adjoint code for a given cost function. However,
governing equations of the ow eld are introduced there is on-going research at Rice University to de-
as a constraint in such a way that the nal expres- velop ADJIFOR [9] which automatically generates
sion for the gradient does not require reevaluation of a discrete adjoint code from existing analysis soft-
the ow eld. In order to achieve this, w must be ware. So far, however, this has not realized the same
eliminated from the above equation. Suppose that level of eciency. In the present work, the adjoint
the governing equation R, which expresses the de- equations have been derived analytically and then
pendence of w and F within the ow eld domain D, approximated in discrete form.
can be written as
R(w; F ) = 0: (1)
3.2 Search Procedure
The remaining cost issue is related to nding a loca-
Then w is determined from the equation tion in the design space where the gradient vanishes,
 @R   @R  and hence there is a local optimum. Normally, this
R = @w w + @ F  F = 0: search starts from a baseline design and the design
space is traversed by a search method. The nal state
Next, introducing a Lagrange multiplier , we have of the search may be subject to constraints imposed
T T      on the design space, yet there is no requirement that
@I @I
I = @w w+ @ F  F T @R @R the trajectory adhere to these except at its end point.
@w w + @ F  F : The eciency of the search depends on the number
of steps it takes to nd a local minimum as well as
With some rearrangement the cost of each step.
 @I T  @R   @I T  @R  In order to accelerate the search, one may resort to
I = @w T @w w+ @ F T
@F  F : using the Newton method. Here, the search direction
is based on the equation represented by the vanish-
Choosing to satisfy the adjoint equation ing of the gradient, G (F ) = 0, and is solved by the
 @R T T standard Newton iteration for nonlinear equations.
@I
= @w (2) Suppose the Hessian is denoted by
@w
the term multiplying w can be eliminated in the A = @@FG
variation of the cost function, and we nd that then the result of a step  F may be linearized as
I = G  F ; G (F +  F ) = G (F ) + A F

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 6 of 26

This is set to zero for a Newton step; therefore
F = A 1G
The Newton method is generally very e ective if the
Hessian can be evaluated accurately and cheaply.
Unfortunately, this is not the case with aerodynamic
shape optimization.
Quasi-Newton methods estimate A or A 1 from
the changes of G recorded during successive steps.
For a discrete problem with N design variables, it
requires N steps to obtain a complete estimate of
the Hessian, and these methods have the property
that they can nd the minimum of a quadratic form
in exactly N steps. Thus in general, the cost of a
quasi-Newton search scales with the dimension of the
design space.
Ecient aerodynamic shapes are predominately
smooth. This suggests a natural alternative ap-
proach to the search method. In order to make sure
that each new shape in the optimization sequence
remains smooth, one may smooth the gradient and
replace G by its smoothed value G in the descent pro-
cess. This also acts as a preconditioner which allows
the use of much larger steps. To apply smoothing in
the 1 direction, for example, the smoothed gradient
G may be calculated from a discrete approximation
to
G @@  @@ G = G (3)
1 1
3.3 Computational Costs
In order to address the issues of the search costs,
the authors investigated a variety of techniques in
Reference [10] using a trajectory optimization prob-
lem (the brachistochrone) as a representative model.
The study veri ed that the search cost (i.e., number
of steps) of a simple steepest descent method applied
to this problem scales as N 2 , where N is the number
of design variables, while the cost of quasi-Newton
methods scaled linearly with N as expected. On the
other hand, with an appropriate amount of smooth-
ing, the smoothed descent method converged in a
xed number of steps, independent of N . Consid-
ering that the evaluation of the gradient by a nite
di erence method requires N + 1 ow calculations,
while the cost of its evaluation by the adjoint method
is roughly that of two ow calculations, one arrives
at the estimates of total computational cost given in
Tables 1-2.
Table 1: Cost of Search Algorithm.
Steepest Descent O(N 2 ) steps
Quasi-Newton O(N ) steps
Smoothed Gradient O(K ) steps
(Note: K is independent of N )

where  is the smoothing parameter. Then, if one sets Table 2: Total Computational Cost of Design.
 F = G, assuming the modi cation is applied on Finite Di erence Gradients
the surface 2 = constant, the rst order change in + Steepest Descent O(N 3 )
the cost function is Finite Di erence Gradients
ZZ + Quasi-Newton Search O(N 2 )
I = G  F d1d3 Adjoint Gradients
ZZ  + Quasi-Newton Search O(N )
 @ @ G  Adjoint Gradients
=  G @  @ G d1d3 + Smoothed Gradient Search O(K )
 @ G 2 !
1 1
ZZ (Note: K is independent of N )
=  G2 +  @ d1d3
1
< 0: We also investigated some other methods which
further improved the dimension-independent conver-
Thus, an improvement is assured if  is suciently gence rate, including multigrid and postconditioning
small and positive, unless the process has already with a Krylov subspace acceleration. Implementing
reached a stationary point at which G = 0 (and there- these in the current aerodynamic shape optimization
fore, according to Equation 3, G = 0). software consistently converges the design to a lo-
It turns out that this approach is extremely toler- cal optimum within 30-60 steps, even for problems
ant to the use of approximate values of the gradient, with thousands of design variables. Moreover, be-
so that neither the ow solution nor the adjoint solu- cause they do not require either the ow or adjoint
tion need be fully converged before making a shape solutions to be fully converged, complete optimiza-
change. This results in very large savings in the com- tions are routinely completed with a computational
putational cost of the complete optimization process. cost equivalent to 2-10 converged ow solutions. As

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 7 of 26

a consequence, our standard practice is to allow ev- tions, both the ow and the adjoint equations must
ery discrete surface point within the CFD grid to be be discretized. The control theory might be applied
its own design function, aligned with the grid line directly to the discrete ow equations which result
emanating from the surface. The amplitude of this from the numerical approximation of the ow equa-
design variable corresponds to the signed distance tions by nite element, nite volume or nite dif-
from the original baseline surface node. This typi- ference procedures. This leads directly to a set of
cally leads to design space dimensions of N > 4000 discrete adjoint equations with a matrix which is
for three-dimensional aerodynamic shape optimiza- the transpose of the Jacobian matrix of the full set
tions. of discrete nonlinear ow equations. On a three-
With the exibility of not being constrained by dimensional mesh with indices i; j; k the individ-
the number of design variables that one can use, thisual adjoint equations may be derived by collecting
has an added bene t that the optimization software together all the terms multiplied by the variation
can be written in such a manner that the end user is wi;j;k of the discrete ow variable wi;j;k. The re-
not burdened with the task of de ning a set of shape sulting discrete adjoint equations represent a possible
functions. A common practice is to specify shape discretization of the adjoint partial di erential equa-
functions of the bump-function class which must be tion. If these equations are solved exactly they can
tailored for each speci c application. The main rea- provide the exact gradient of the cost function which
son for using bump functions is to reduce the number results from the discretization of the ow equations,
of design variables needed and yet obtain reasonable which is itself, however, inexact. On the other hand
results. Specifying an appropriate set of bump func- any consistent discretization of the adjoint partial
tions for a given problem is somewhat of a black art di erential equation will yield the exact gradient in
in itself, and getting an e ective set frequently re-the limit as the mesh is re ned.
quires experimentation by the user. The process is There are a number of bene ts to be gained from
also prone to input error that may not be discovered developing the theory for the partial di erential
until an optimization run fails to produce reasonableequations of the ow. First, the true optimum shape
results. This is a poor environment for both user belongs to an in nitely dimensional space of design
and design, as the "optimized" design will only be asparameters, and the theory provides an indication, in
good as the user is at choosing his shape-function set
principle, of how such a solution could be approached
and accurately inputting it. Hence, the need for an if sucient computational resources were available.
expert user is mandated. It is the authors' position Second, it provides insight into the nature of the
that optimization software should free the engineer adjoint equations, and the connection between the
from mundane tasks and allow him to focus on the formulation of the cost function and the boundary
more global requirements of the system development. conditions needed to assure a well-posed problem.
We acknowledge that our software has not reached Third, in certain circumstances the discrete solution
this level of usability; yet there exists a path to get
may lose the property of continuous dependence of
there. the design parameters. It may, for example, con-
In some of our applications where the satisfactiontain non-di erentiable ux limiters. Also, if adaptive
of constraints was important, the adjoint method was mesh re nement is used, there will be a discontinu-
used to provide sensitivity information to an external
ous change in the solution whenever a mesh point is
optimization method such as NPSOL [11]. NPSOL added or deleted. Finally, the di erential equation
implements a sequential quadratic programming al- theory provides a guideline for the design of itera-
gorithm and allows for the consideration of both lin-tive solution methods for the adjoint equation, both
ear and non-linear constraints. However, the overall in the case when the adjoint equation is separately
cost of the optimization procedure was greatly in- discretized and in the case when the discrete adjoint
creased, because of the need to conduct line searchesequations are derived directly from the discrete ow
every time a design change direction is chosen. equations. The theory for standard multigrid meth-
ods, for example, depends on the property that the
3.4 Continuous vs. Discrete Adjoint sent the equations
discrete
same di
on a sequence of meshes all repre-
erential equation. It turns out that
Formulation the same multigrid solution method can readily be
In reference [12] the rst author derived the adjoint used for both the ow and the adjoint equation.
equations for transonic ows modeled by both the A number of researchers have suggested that it
potential ow equation and the Euler equations. The is necessary to use the discrete adjoint to avoid an
theory was developed in terms of partial di eren- inconsistency with the discrete cost function. Other-
tial equations, leading to an adjoint partial di er- wise the resulting "noise" in the search process may
ential equation. In order to obtain numerical solu- prevent full convergence of the search. However in

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 8 of 26

our work, we avoid this inconsistency by driving the
search process exclusively with the value of the com-
puted gradient without ever using the value of the
cost function. The search is terminated when the
gradient is suciently close to zero. We also ad-
dressed this issue in our study of the brachistochrone
problem [10]. In this case, one can discretize the an-
alytic form of the gradient derived by the Calculus
of Variations, which corresponds to the use of the
continuous gradient in the aerodynamic shape opti-
mization problem. Alternatively, one can derive a
discrete gradient directly from the discrete integral
of the cost function. Our studies showed that the
use of the continuous gradient consistently yielded a
slightly more accurate result, at least for this partic-
ular problem. It should be noted that for methods
requiring line searches, instead of identifying the lo-
cation where the cost function is a minimum, one
should terminate the line search at the point where
the search direction and the computed gradient are
orthogonal.
4 Case Studies
Two case studies are included here to illustrate the
impact of our aerodynamic shape optimization meth-
ods on the design of transonic wings. The rst is on
the design of a concept race plane for the annual
Reno air races. The second is on the optimization
of the British-Aerospace MDO datum wing. Other
design e orts which have utilized these methods in-
clude: Raytheon's business jets, NASA's High-Speed
Civil Transport, regional jet designs, as well as sev-
eral Boeing projects such as the Blended-Wing-Body
and the MDXX.
4.1 Reno Race Plane
The design objectives of the Reno racer are given
with respect to a standard day at the race location
which is at 5000' MSL and ISA +20 C. The top
speed in straight and level ight is to exceed 600
MPH TAS. The average lap speed around the 1999
unlimited race course is to exceed 550 MPH TAS.
The aircraft is to be capable of sustaining a 9G ma-
neuver load, subject to a 5G gust load; yielding a 14G
limit load with a 1.5X safety factor. Roll rate should
exceed 200 per sec at 350 KEAS. Stall speed should
be less than 90 KEAS. Landing distance should not
exceed 1500', dead stick. Note that some of these re-
quirements are more stringent than the performance
of some state-of-the-art jet ghters.
The design requirements state that the aircraft
must be piston powered and propeller driven. Engine
power-to-weight ratio for reliability at continuous
Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV
output should be about 2.5 HP/lb. for a turbo-
charged piston engine with gear reduction and other
accessories. The stability and control is to be pro-
vided by a manual, unboosted system with posi-
tive static and dynamic margins that exceed current
unlimited-class race planes. There should be min-
imal change in stability between power on and o .
For crew provisions, the design allows dual pilots in
a tandem seating arrangement, with seats inclined
30 for G tolerance, and include MIL-SPEC oxygen
and G-suit connections. Low altitude ejection for
both pilots is also required.
The development of this aircraft began from the
ground up, as an all-new design. Every major ele-
ment of the airplane had to be engineered. This in-
cluded the airplane's general layout, a unique propul-
sion system, the aerodynamic designs of the wing,
fuselage and empennage, as well as the ecient in-
tegration of these and other subsystems. Although
a unique propulsion system eventually became our
baseline design, several systems were considered.
These included a conventional tractor propeller with
a front-mounted engine, and two mid-engine designs
{ one with a pusher prop aft of the tail and the other
a body-prop design. While a tractor design is much
more conventional, the design requirements favored
a mid-engine concept. Avoiding propeller strike for
the pusher design during rotation was a major issue.
While there were many other factors that played into
our decision, the body-prop design became our base-
line con guration. With the propeller mounted aft of
the wing, this concept also provided the possibility of
promoting laminar ow on the forward fuselage and
wing surfaces. A side view of the body-prop's general
layout is provided in Figure 5 and a computer graph-
ics rendering of this con guration in ight is given in
Figure 6. The highest risk item of this design is de -
nitely related to engineering the structure to accom-
modate the load path between the tail and center
wing box.
One can see from the general layout that the ver-
tical tail (rudder) is rigged downward instead of in a
normal upward position. This was done for two rea-
sons; rst to provide a skid at the rudder tip to pre-
vent propeller strike, and second to keep it in clean
air during a high, positive G maneuver.
The complete aircraft design e ort has been con-
ducted by a very small team. The program manager
is Eric Ahlstrom of Star Aviation, Robert Gregg is
the aerodynamic lead, and in addition to the authors,
Dennis McDowell and Mark DeHaan complete the
aerodynamics team. The authors were tasked with
the responsibility to design the outer-mold-line sur-
faces of the wing and fuselage components.
During this multi-disciplinary design e ort, the
general layout of the body-prop concept race plane
9 of 26
evolved as the design team better understood how for very high wing loading, the stall speed require-
to maximize the performance of the integrated sys- ment sized the wing area. The maneuver loads, sus-
tem. Normally, global changes such as those encoun- tained turn rate, and gust loads required that the
tered are very disruptive during the design of a high- wing have no bu et at CLTotal = 0:64; M = 0:72. A
performance, transonic wing. However, utilization trade study of wing thickness, sweep and taper ratio
of the aerodynamic shape optimization software de- was made using NACA SC(2) airfoils as a baseline.
veloped by the authors allowed various aircraft sub- From this, a section thickness of 13.5% at the wing
systems to be routinely modi ed without adversely root and 12% at the tip was chosen, combined with
impacting development costs or schedule; new wing a quarter-chord sweep of 28 to meet the Mdd = 0:8
designs occurred over night. Our ability to perform requirement. An aspect ratio of AR = 8:3 and a
new optimizations over night, on a ordable comp- taper ratio of  = 0:45 were chosen to allow a wing-
uters, was a key factor which allowed this form of tip extension for a growth airplane. Conversely, a
simulation-based aerodynamic design work to be em- production break was included at 87% semi-span to
braced by the rest of the design team. The complete allow a 4ft2 reduction in wing area if ever needed. A
evolution of the aircraft's general layout was accom- planform Yehudi (inboard chord extension) was in-
plished in a very compressed time frame: our aero- corporated into the wing trailing edge to accommo-
dynamic shape optimizations played a pivotal role in date the main landing gear. Inclusion of this Yehudi
this achievement. More importantly, this evolution also helped reduce the wing downwash angle of the
was required to meet all of the design goals imposed ow entering the propeller. The wing is a two-spar
on the team by our sponsor. For more detailed in- design with spars at 15% and 65% chord, and is aug-
formation on this aircraft design, see Reference [13]. mented with a secondary spar behind the main gear
wells that parallel the Yehudi trailing edge. This sec-
Wing Design ondary spar provides structural support at the main
gear pivots. A one-piece wing box construction will
The design of the wing geometry occurred in several be used to reduce weight and complexity.
phases; the duration of each of the rst ve phases
lasted from 1 day to 1 week long. In most cases, there
was a lapse between phases, as time was required for 4.1.2 Phase II: Rough Detailed Design
the team to digest the evolution of the aircraft design
The baseline wing of phase II was de ned using air-
and formulate new ideas to investigate. foil sections derived from NACA 64 sections, scaled
to conform to the planform and thickness distribu-
4.1.1 Phase I: Conceptual Layout tion established in phase I. Some cursory 2D aero-
The basic requirements de ned in phase I were dynamic optimizations were performed on these sec-
based on conceptual methods and design charts. tions to better tailor their characteristics for the ini-
These requirements included the general layout of tial design conditions; the 2D conditions and geom-
the wing (planform & thickness distribution), the etry transformations used for this e ort were based
design cruise condition on simple-sweep theory. SYN103 was run in Euler,
drag-minimization mode for this 2D design e ort.
M = 0:77; CLTotal = 0:32; Ren = 14:5M; The remaining unspeci ed geometric quantity for the
wing was its twist distribution. To set this, FLO22
the o -design capabilities for bu et was used to provide the span load of the wing. This
CLBuffet = 0:64 at M = 0:72; code, which solves the three-dimensional transonic
potential ow equation, has been extensively used
the drag divergence Mach number since its inception in 1976. (For reference, FLO22
runs in about 5 seconds on an AMD Athlon 850
Mdd = 0:80 at CL = 0:1; MHz PC.) While FLO22 is a wing-only CFD code,
the clean wing maximum lift coecient pseudo-fuselage e ects were included in the present
work. The rst pseudo-body in uence is its accel-
CLmaxCW  1:6 at M = 0:2; eration of the on-set Mach number at a critical sta-
tion on the wing; typically this is around 50%-60%
and a pad on the divergence Mach number to allow semi-span. Running the isolated fuselage geometry
room for growth in out years. in a surface-panel method and interrogating the ow-
The conceptual methods set the wing area, eld velocity at the critical wing station determines
Sref = 75ft2 , to provide a wing loading range of this acceleration. The second pseudo-body e ect is
40 60 lbs:=ft2 . Despite the high dynamic pres- how the presence of the fuselage at an angle of at-
sures of the racing environment and the opportunity tack warps the ow eld's local angle of attack as a

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 10 of 26

function of span location. The third pseudo-body in-
uence is related to the carry-over lift of the wing's
circulation onto the fuselage. This ratio is de ned as
CLTotal =CLWing , is 1:22 for this con guration and
was determined by running a surface-panel method
on the wing/body combination. These pseudo-body
e ects are included in FLO22's wing-only solution by
running the exposed wing in the code at the wing's
CL, at a higher Mach number and re-referencing the
results back to the original Mach, and adding a delta-
twist distribution to the wing to simulate the ow-
eld warping. Using this procedure, a twist distri-
bution was speci ed that yielded a near-elliptic span
loading. This initial design was done very rapidly,
covering only a two-day period, and provided a point
to start the 3D design e ort.
The initial FLO22 analyses indicated that the
wing design requirements could be satis ed; the
initial wing had a Mach capability of 0:775 at
CLTotal = 0:3. However, there was serious concern
with the body e ects of the fuselage's low neness
ratio. The team was relatively sure that the baseline
wing would have problems near the root region be-
cause of the atypical contouring of the fuselage geom-
etry.
4.1.3 Phase III: Aero Optimization
In phase III, the rst step was to assess the issues
existing with the baseline wing geometry, designated
Shark1, as it integrates with the fuselage. This anal-
ysis was performed using SYN88, and is illustrated
in Figure 7. SYN88 is a wing/body Euler method
which also incorporates an adjoint-based optimiza-
tion procedure for aerodynamic shape design. In
Figure 7, pressure distributions at several stations on
the wing are provided. Adhering to standard aerody-
namic practices, the pressure coecient of the sub-
plots are presented with the negative axis upward.
The area trapped by the upper- and lower-surface
pressure-distribution curves is equivalent to the lo-
cal sectional lift coecient. Each subplot is linked
graphically to its corresponding location on the wing
depicted in the center of the gure. Also included on
the wing-planform plot are the upper-surface isobars
of the rst solution which is depicted by the solid
curves in the perimeter subplots. A shock is evi-
dent with a concentration of contour lines in the iso-
bar image and corresponds to a sharp discontinuity
in the pressure-distribution subplots. The quanti-
ties in the legend of this gure correspond to the
wing forces. The drag listed is only the inviscid
drag (induced+shock). Recall that the design lift
was CLTotal = 0:32 and that the carry-over lift ratio
was 1:22 for this con guration. Hence, the wing lift is
CLWing = CLTotal =1:22 = 0:27. The other item to
Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV
note is that this analysis was performed at a Mach
number of M = 0:78 rather than M = 0:77. The
reason for this increase in freestream Mach number
was to allow for the acceleration of the ow eld near
the wing root from the propulsion system. Meth-
ods based on actuator-disc and blade-element theo-
ries determined this acceleration to be
M ' 0:01.
Since the wing-root region was of utmost concern at
this stage in the design, the full level of propeller
e ects on the on-set Mach number was used. Re-
ferring to the wing-planform plot of Figure 7, notice
the strong shock that unsweeps as it nears the side-
of-body. The main purpose of a swept-back wing is
to reduce the normal Mach number of the ow into
a shock, however, if the shock unsweeps, this bene t
is lost. As suspected, the contouring of the fuse-
lage cross-sections had an adverse e ect on the wing
aerodynamics, and unfortunately it was worse than
expected. The inviscid drag (induced+shock) of the
wing was CDWingINV = 180 counts for the baseline
con guration.
Phase III continued by running SYN88 in drag-
minimization mode, constraining the wing modi -
cations to be thicker everywhere than the baseline
geometry to maintain structural depth; the fuselage
geometry was frozen. Initially, these were single-
point optimizations at the 4G design condition, just
to scope the potential bene t. (For reference, a
SYN88 wing/body analysis takes about 10 minutes
on a Sony Vaio notebook computer with a 750 MHz
Pentium II chip; a single-point optimization takes
about 50 minutes using a mesh with 256 x 32 x 48
cells in a C-H topology, for a total of 393; 216 cells.)
Eventually, all optimizations were migrated to triple-
point designs that considered a range of lifting con-
ditions at the design Mach number. This range cor-
responded to variation and persistence of G loads
being pulled during a lap of the race course. The
design Mach number corresponded to an average
speed around the track.
In the rst optimization run, SYN88 reduced the
wing's inviscid drag from 180 counts to 104 counts
in 30 design cycles. The results of this optimiza-
tion are illustrated in Figure 8. Although fairly large
improvements were realized, a strong normal shock
at the side of fuselage could not be eliminated, and
we felt we could do better if the fuselage contour
near the wing trailing edge was allowed to be mod-
i ed. Several concurrent changes to the aircraft's
general layout were being considered. The team was
forming new ideas as the complete system integration
was beginning to be better understood. The changes
that were directly related to the wing design were
the fuselage reshaping and a trailing-edge planform
blending that would allow more room for stowing
the landing-gear structure. The planform modi ca-
11 of 26
tions were made to the current wing and three addi- Tollmien-Schlichting waves would decay rather than
tional fuselages were de ned that stretched it by 1, amplify. The possibility of having runs of laminar
2 and 4 feet aft of the wing-root mid-chord and con- ow was achievable. The dilemma, however, was
sistent with the engine packaging requirements. In could laminar ow be achieved in the eld? The pri-
fact, the trailing-edge modi cation was also done in mary mission of this plane occurs just above ground
a manner to help alleviate the shock-unsweep prob- level where bug strikes are sure to occur, thus con-
lem, as well as accommodate the landing gear. This taminating the wing's leading edge. We decided to
planform change proved to be bene cial as another investigate whether or not the wing's pressure distri-
triple-point drag minimization was performed, which butions could be tailored to have favorable gradients
dropped the wing's inviscid drag from 148 counts to for up to 40% chord without adversely a ecting the
98 counts at the design point with the original fuse-aerodynamic performance of the fully-turbulent wing
lage. Repeating similar triple-point optimizations ondesign. If it could, then the resulting design would be
1-, 2- and 4-foot fuselage extensions provided su- adopted, yet without taking credit for laminar- ow
cient data to show that the shock unsweep problem drag reductions in our performance estimates.
could be completely eliminated with a 2-foot fuse- Phase IV concentrated on promoting laminar ow
lage stretch. This optimization reduced the wing's on the wing without degrading the performance of
inviscid drag from 92 counts to 74 counts within 30 the wing if it was fully turbulent as compared with
design cycles; the resulting wing geometry was desig-the fully-turbulent design of Shark52. This objective
nated Shark52. The pressure distributions and drag was not limited to the design point, but rather was
loops for Shark52 at M = 0:78 and CLWing = 0:27 expanded to include a Mach number range M  0:74
are shown in Figures 9-10. and a lift range of CLWing  0:27. The rst task
It should be emphasized that within the course was to compute the viscous ow about the Shark52
of one week, the wing geometry had evolved from con guration at various ow conditions. This was
Shark1, which produced 180 counts of inviscid drag, accomplished using SYN107P, a wing/body Navier-
to Shark52, which only had 74 counts at the design Stokes method for analysis and design. The design
point. During this week, the wing planform changed Reynolds number was Re = 14:5M , based on the
and the fuselage length stretched. This is an ex- reference chord. (For reference, SYN107P runs in
tremely large improvement which was accomplished parallel under MPI; on a 16 processor AMD Athlon
in a very compressed time! Furthermore, the 650 MHz cluster, an analysis takes about 30 min-
database of CFD solutions O(100) had grown large utes of wall-clock time, while an optimization takes
enough that very informed modi cations to the con- about 2 hours.) Starting with the computed pressure
guration could be made. These included a wing- distributions of Shark52, a series of inverse designs
planform change to better stow the landing gear, as were performed, also with SYN107P. It was easier to
redesign the wing at the higher Mach number and
well as the fuselage extension to eliminate the normal
shock. Figure 10 shows a side-by-side comparison of accommodate the requirements at M = 0:74, rather
the pressure isobars of the Shark52 wing and the ini-than the other way around. This study was com-
pleted with the wing geometry designated SharkNS7.
tial wing that clearly illustrates the reduction of the
shock strength across the entire wing. This improve- Figure 12 illustrates the pressure distributions for
ment was a result of all of these important changes SharkNS7 at M = 0:78 and a lift range of
to the con guration. 0:18  CLWing  0:34:
The nal aspect of phase III was to rebalance the
At the design point CDWing = 128 counts which
aircraft. This required a 6-inch fuselage stretch for-
ward of the wing to compensate for the 2-foot stretchis composed of CDWingForm = 77 counts and
of the after-body. Once the fuselage geometry was CDWingSkinFriction = 51 counts.
frozen, the wing pressures were polished by running It can be seen that favorable pressure gradients ex-
SYN88 in inverse-design mode. A drag build-up of ist on both upper and lower surfaces through this lift-
this design showed that, at 550 MPH, the maxi- ing range for about the rst 30%-40% chord, depend-
mum L/D of the complete aircraft was estimated to ing on span location and lifting condition. Also note
be 14:78. It occurs at CLTotal = 0:49, which corre- that a very weak shock forms as the lift coecient is
sponds to a 7G maneuver. increased above the design point of CLTotal = 0:26.
On the upper surface the shock will trigger transition
4.1.4 Phase IV: Laminar Flow provided any attachment line contamination from
the fuselage boundary layer is removed by a notch-
The team began to kick around the idea of a laminar- bump, and Re < 200. Re varied from approxi-
ow design. Quick calculations on the attachment- mately 125 just outboard of the fuselage to around
line Reynolds number, Re , indicated that the 80 at the wing tip. The amount of laminar run on

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 12 of 26

the upper surface increases as Mach increases due
to the shock moving aft on the airfoil as well as the
pressure gradients becoming more favorable. At race
conditions, the wing should have an appreciable ex-
tent of laminar ow, provided the surface of the wing
is smooth and free of particulate contamination. The
estimated bene t of the laminar ow runs is between
10 and 20 counts of drag reduction, depending on
Mach number. If this level of drag reduction is ac-
tually realized, it will increase the aircraft's perfor-
mance by an additional 5%, a signi cant improve-
ment.
4.1.5 Phase V: Final Touches
The rst task of phase V was to establish an ap-
propriate leading-edge radius distribution, tailored
for low-speed characteristics, without really changing
the wing pressure distributions at the cruise design
conditions. This modi cation was accomplished with
local, explicit geometry perturbations. An additional
modi cation to the wing thickness distribution was
also done. After these changes were incorporated
into SharkNS7, the nal wing was analyzed to ver-
ify that these geometry changes did not adversely
e ect the pressure distributions. When overlaid on
the same plot, the curves nearly appeared as one.
Finally, clean wing CLmax , CLmaxCW , was com-
puted to ensure the wing satis ed the required clean
wing stall speed. The design requirement was to
provide CLmaxCW > 1:6 at M = 0:2. This was
determined by nding the ow condition where the
wing's Cpmin distribution reached an empirically-
determined critical value. The nal wing provided
a CLmaxCW = 1:64, just meeting the requirement.
The nal wing geometric characteristics are shown
in Figure 13 which illustrates the half-thickness and
camber distributions of the root and outboard airfoil
sections. Note that the root airfoil is 13% thick @
31.5% chord and has -1.2% camber. The outboard
airfoil is 11.5% thick @ 38.6% chord and has +1%
camber. While these thicknesses are about 0.5%
thinner of that speci ed in the conceptual design
stage, it was the team that reduced this thickness,
not the optimization exercises.
4.1.6 Phase VI: Test Evaluation
Phase VI is on-going. The construction of this air-
craft is to begin later this year, and a ight test will
follow. Prior to rst ight, wind-tunnel data will be
collected on the nal wing/body con guration. Fab-
rication of the wind-tunnel model has begun at the
Swedish FFA under a cooperative agreement.
Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV
4.2 BAE MDO Datum Wing
The second case study that we present is not a part
of a complete aircraft design e ort, but rather, rep-
resents a pure aerodynamic optimization. It is in-
cluded to illustrate the level to which the elapsed
time associated with the inner loop of Figure 2 can
be compressed. This exercise was conducted by the
rst author through a request from Murray Cross
of British Aerospace - Airbus Division. In particu-
lar, this e ort was performed using a mesh provided
by Mr. Cross which was generated to be suitable
for Navier-Stokes simulations. The objective of this
study was to minimize the wing's drag at a xed lift-
ing condition, while maintaining the thickness of the
baseline wing, everywhere. This redesign was accom-
plished in 60 design cycles of SYN107. On a Beowulf
cluster comprised of 16 AMD Athlon 650 MHz chips,
an optimization of this class can be completed within
2-3 hours of wall-clock time.
Figures 14-15 illustrate the computed pressure dis-
tributions of the baseline MDO datum wing with
variations in lift and Mach number, respectively, cen-
tered about the cruise design point of M = 0:85 and
CL = 0:45. According to this simulation, the base-
line wing's cruise-point drag is 162.8 counts. In these
gures, note the wavy upper-surface pressure distri-
butions of the baseline wing near its 75% semispan
station.
After the redesign, SYN107 reduced the wing's
drag to 156.8 counts. This 6-count drag reduction
represents about a 2.5% improvement in aircraft per-
formance. For a commercial transport aircraft, this
is a very signi cant gain.
Figures 16-17 are the equivalent of Figures 14-15,
however for the redesigned wing. Note that the new
pressure distributions have a roof-top shape forward
of the shock, and that the wavy character near 75%
semispan has been greatly reduced. Figure 18 com-
pares the optimized and baseline wing pressures at
the design-point ow conditions. One can see that
the shock strength of the optimized wing has been
reduced with respect to that of the original geom-
etry. Figure 19 illustrates that the spanload of the
redesigned wing is essentially the same as that of the
baseline wing. Hence, the drag improvement realized
by the optimization was not a result of moving the
loading further outboard to reduce the vortex drag.
Figure 20 provides a comparison of the airfoil
geometry at the 37% semispan station, before and
after the redesign. In this gure, the original airfoil
is depicted by the solid line while the redesign is illus-
trated with the dashed line. Note that the thickness
of the redesigned airfoil is greater than that of the
baseline section. It is also interesting to note that
camber level has been reduced along the chord.
13 of 26
 This study at this stage actually represents only
the beginning of an aircraft design. To continue
with a real aircraft design, the designers would scru-
tinize the changes in the geometry made by the aero-
dynamic shape optimization and attempt to gar-
ner what this synthesis was attempting to accom-
plish and where it was being held back through the
prescribed constraints. For example, we have en-
countered cases where the wing thickness outboard
grew away from the constraints, while at the same
time was being limited in the region of the planform
break. Upon further investigation in the particular
case, we found that the provided limits in this re-
gion were simply the result of interpolating hard con-
straints near the side-of-body with hard constraints
for the outboard wing. Understanding that the origi-
nal thickness constraint at the planform break was an
"arti cial" one, we asked the design team to better
de ne what the hard constraint in this region should
be. Consequently, another optimization provided in-
creasingly better aerodynamic performance.
5 Conclusions for the Present
As can be seen from the case studies, aerodynamic
shape optimization can signi cantly streamline the
design process. The present method has now been
successfully applied in a variety of projects, including
the McDonnell-Douglas MDXX, the NASA HSCT
studies, the Boeing Blended-Wing-Body project, and
the Beech Premier.
In fact, there are many di erent examples on how
aircraft design teams have utilized the rapidly pro-
vided information of aerodynamic shape optimiza-
tion to make improvements to their aircraft con g-
urations. The diversity of these examples illustrate
the artistic and creative nature of the thought pro-
cesses by the design teams. It is through these unpre-
dictable paths in design direction that dramatic im-
provements of the multi-disciplinary systems are ac-
complished. Further, because unforeseen directions
may be required to accomplished the design goals, we
feel that it is highly unlikely that the designers will
be replaced by a comprehensive, multi-disciplinary
optimization (MDO) method. On the other hand,
there are very well established dependencies between
di erent disciplines that can and should be coupled
for MDO.
Although the search method is only guaranteed to
nd a local minimum, it turns out in practice that
our aerodynamic optimizations are yielding results
that are in the neighborhood of the known lower
bounds for aerodynamic drag, as determined by opti-
mum span loading, at-plate skin-friction, and min-
imum wave drag. Furthermore, in practice, the de-
Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV
signer's goal is not to determine the absolute best
design, but rather, is tasked to make the most im-
provement to a design in a xed amount of time spec-
i ed by program schedule.
Our method is not intended to replace the judge-
ment and insight of the aircraft designers. Rather, it
should properly be viewed as an enabling tool that
allows the designers to focus their e orts on the cre-
ative aspects of aircraft design, by relieving them of
the need to spend large amounts of time exploring
small variations. By intelligent choice of the cost
function to measure the aerodynamic performance
and perhaps also the deviation from a desired pres-
sure architecture, one can essentially eliminate the
need to carry out detailed section design. Instead,
the designers can concentrate their attention on large
scale parameters such as wing span, area and sweep,
knowing that the optimization process will realize
the best possible performance for any given choice of
these parameters.
We plan to extend the software to allow for plan-
form variations in the optimization. However, these
involve multi-disciplinary trade-o s, such as that be-
tween increased structure weight and reduced vortex
drag as the wing span is increased. Also variations,
for example, in sweep-back angle may require the
use of multi-objective optimization to take account of
both aerodynamic performance and handling quali-
ties. Multi-objective optimization is also likely to be
needed for the aerodynamic design of military air-
craft. We envision that it may pay to embed the
existing software as an inner loop in a global optimi-
zation process, where it may be used, for example,
to facilitate the determination of Pareto fronts for
con icting objectives.
A multi-block implementation of the software has
been successfully used for the design of complete
con gurations [14, 15, 16, 17]. However, the ex-
isting problems of geometry modelling, CAD repair
and grid generation present a major bottleneck in
the design process. In order to alleviate this prob-
lem, we are currently in the process of implementing
an extension of the method to unstructured meshes
of arbitrary elements. Our goal is to produce a
"mesh-blind" scheme, which does not need to know
what kind of cells are contained in the mesh, but
which should retain the computational eciency of
the present structured-mesh method.
6 Visions for the Future
Looking into the future, it may be anticipated that
increasingly elaborate simulations will be performed
to support the design process as greater computing
power becomes available at very a ordable costs. It
14 of 26
is likely that large eddy simulation (LES) will eventu-
ally be used to study unsteady and separated ows,
such as those associated with high-lift systems. On
the other hand, it may be noted that aerodynamic
design can be brought to an advanced level of so-
phistication, probably quite close to the maximum
realizable cruise performance, using calculations of
the complexity presented in this paper.
Hence, it is interesting to consider the impact of
being able to carry out calculations of a xed size
with progressively smaller and cheaper equipment.
The Sony Vaio 505 superslim laptop weighs 3.75
pounds and costs about $3,000. Yet it has process-
ing power, random access memory and disk capac-
ity comparable to that of the Convex C2 of 1985,
which weighed about 1,000 pounds and cost about
$600,000. These weight and cost reduction ratios are
about 200 to 250, and have occurred over the past
15 years. The following graph illustrates this trend.
Convex
61000 lbs
$600K
Sony
HH 3.75
Log HHH $3Klbs Wrist Watch
HH
#,$ HHH Computer
HHH0.25
$15
oz
jH
- and P. Hovland. Generating derivative codes
1985 2000 2015
Impact of Computer Evolution
with Constant Performance
If this trend continues unabated for the next 15 years,
which may be possible before a molecular barrier to
microprocessor or circuit density is reached, then we
should be able to carry out both ow simulations
and aerodynamic design optimization on a machine
the size of a wrist watch, with a cost on the or-
der of $15. It is likely that voice and hand-writing
recognition will eventually replace the keyboard and
mouse as the primary interaction between the user
and computer. Miniturized virtual screens already
exist which provide the equivalent of a 60-inch dis-
play which appears to oat about 8 feet in front of the
user. These virtual screens can service each eye inde-
pendently such that stereoscopic visualization will be
the norm. Wireless communication will be used for
large-scale data I/O, as well as to maintain constant
connectivity amoung design-team members. Clearly,
this emphasizes the crucial need to devise more e-
cient methods for geometry de nition, mesh genera-
Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV
tion, and post processing. It seems certain, however,
that it will be possible to carry out sophisticated
aerodynamic design with very small teams, and with
the members not necessarily at the same location.
Designers in such an era will have the opportunity
to fully explore their creative ideas.
Now if only that wrist watch will be water proof
and shock resistant.
Acknowledgment
This work has bene ted from the generous support
of the Air Force Oce of Scienti c Research under
Grant No. AF F49620-98-1-002, as well as the con-
tributions of The Boeing Company. We are greatful
to British Aerospace, Airbus Division for releasing
their MDO datum wing. Renaissance Research pro-
vided partial funding for the Reno Race Plane design
e ort.
References
[1] M. Van Dyke. An Album of Fluid Motion. The
Parabolic Press, Stanford, 1982.
[2] R. M. Hicks and P. A. Henne. Wing design
by numerical optimization. Journal of Aircraft,
15:407{412, 1978.
[3] C. Bischof, A. Carle, G. Corliss, A. Griewank,
from Fortran programs. Internal report MCS-
P263-0991, Computer Science Division, Ar-
gonne National Lab. and Center of Research on
Parallel Computation, Rice Univ., 1991.
[4] L. L. Green, P. A. Newman, and K. J. Haigler.
Sensitivity derivatives for advanced CFD al-
gorithm and viscous modeling parameters via
automatic di erentiation. AIAA paper 93-
3321, 11th AIAA Computational Fluid Dynam-
ics Conference, Orlando, Florida, 1993.
[5] W. K. Anderson, J. C. Newman, D. L. Whit-
eld, and E. J. Nielsen. Sensitivity analysis
for the Navier-Stokes equations on unstructured
meshes using complex variables. AIAA pa-
per 99-3294, Norfolk, VA, June 1999.
[6] A. Jameson, L. Martinelli, J. J. Alonso, J. C.
Vassberg, and J. Reuther. Simulation based
aerodynamic design. IEEE Aerospace Confer-
ence, Big Sky, MO, March 2000.
[7] A. Jameson. Optimum aerodynamic design us-
ing control theory. Computational Fluid Dy-
namics Review, pages 495{528, 1995.
15 of 26
 [8] A. Jameson. Optimum aerodynamic design us-
ing CFD and control theory. AIAA paper 95-
1729, AIAA 12th Computational Fluid Dynam-
ics Conference, San Diego, CA, June 1995.
[9] A. Carle, M. Fagan, and L. L. Green. Prelimi-
nary results from the application of automated
adjoint code generation to CFL3D. AIAA pa-
per 98-4807, 1998.
[10] A. Jameson and J. C. Vassberg. Studies of alter-
native numerical optimization methods applied
to the brachistochrone problem. In Proceedings
of OptiCON'99, Newport Beach, CA, October
1999.
[11] P. E. Gill, W. Murray, M. A. Saunders, and
M. H. Wright. User's guide for npsol (ver-
sion 4.0): A fortran package for nonlinear pro-
gramming. Technical Report SOL 86-2, Depart-
ment of Operations Research, Stanford Univer-
sity, Jan. 1986.
[12] A. Jameson. Aerodynamic design via control
theory. Journal of Scienti c Computing, 3:233{
260, 1988.
[13] E. Ahlstrom, R. Gregg, J. Vassberg, and
A. Jameson. G-Force: The design of an unlim-
ited class Reno racer. AIAA paper 2000-4341,
18th AIAA Applied Aerodynamics Conference,
Denver, CO, August 2000.
[14] J. Gallman, J. Reuther, N. Pfei er, W. Forrest,
and D. Bernstorf. Business jet wing design us-
ing aerodynamic shape optimization. AIAA pa-
per 96-0554, 34th Aerospace Sciences Meeting
and Exhibit, Reno, Nevada, January 1996.
[15] J. Reuther, J.J. Alonso, M.J. Rimlinger, and
A. Jameson. Aerodynamic shape optimization
of supersonic aircraft con gurations via an ad-
joint formulation on parallel computers. AIAA
paper 96-4045, 6th AIAA/NASA/ISSMO Sym-
posium on Multidisciplinary Analysis and Opti-
mization, Bellevue, WA, September 1996.
[16] J. J. Reuther, A. Jameson, J. J. Alonso, M. Rim-
linger, and D. Saunders. Constrained multipoint
aerodynamic shape optimization using an ad-
joint formulation and parallel computers: Part
i. Journal of Aircraft, 36:51{60, 1998.
[17] J. J. Reuther, A. Jameson, J. J. Alonso, M. Rim-
linger, and D. Saunders. Constrained multipoint
aerodynamic shape optimization using an ad-
joint formulation and parallel computers: Part
ii. Journal of Aircraft, 36:61{74, 1998.

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 16 of 26

 Figure 3: Miss Ashley II and Rare Bear en Route.

Figure 4: Reno Race Course Layout.

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 17 of 26

 Figure 5: Side View of Body-Prop Design.

Figure 6: Rendering of Body-Prop Design in Flight.

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 18 of 26

 MACH = .780
-1.5 -1.5
SYMBOL SOURCE ALPHA CL CD
SYN88 - Shark1 1.000 .2678 .01796
Cp -1.0 -1.0

Cp
-0.5 -0.5

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
X/C X/C
0.5 Shark1 0.5
47.8% Span 90.0% Span
Upper-Surface Isobars
-1.5 ( Contours at .05 Cp ) -1.5

-1.0 -1.0
Cp

Cp
-0.5 -0.5

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
X/C X/C
0.5 32.5% Span 0.5 75.7% Span
-1.5 -1.5

-1.0 -1.0
Cp

Cp
-0.5 -0.5

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
X/C X/C
0.5 18.0% Span 0.5 61.6% Span

Figure 7: Pressure Distributions of Shark1 Baseline Wing.

MACH = .780
-1.5 -1.5
SYMBOL SOURCE ALPHA CL CD
SYN88 - Shark5 .706 .2721 .01043
-1.0 -1.0
SYN88 - Shark1 1.000 .2678 .01796
Cp

Cp

-0.5 -0.5

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
X/C X/C
0.5 Shark5 0.5
47.8% Span 90.0% Span
Upper-Surface Isobars
-1.5 ( Contours at .05 Cp ) -1.5

-1.0 -1.0
Cp

Cp

-0.5 -0.5

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
X/C X/C
0.5 32.5% Span 0.5 75.7% Span
-1.5 -1.5

-1.0 -1.0
Cp

Cp

-0.5 -0.5

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
X/C X/C
0.5 18.0% Span 0.5 61.6% Span

Figure 8: Comparison of Shark5 Wing on Baseline Fuselage with Baseline Con guration.

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 19 of 26

 MACH = .780
-1.5 -1.5
SYMBOL SOURCE ALPHA CL CD
SYN88 - Shark52 .718 .2714 .00738
-1.0 -1.0
Cp SYN88 - Shark1 1.000 .2678 .01796

Cp
-0.5 -0.5

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
X/C Shark52 X/C
0.5 47.5% Span 0.5 89.9% Span
Upper-Surface Isobars
-1.5 ( Contours at .05 Cp ) -1.5

-1.0 -1.0
Cp

Cp
-0.5 -0.5

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
X/C X/C
0.5 32.4% Span 0.5 75.4% Span
-1.5 -1.5

-1.0 -1.0
Cp

Cp
-0.5 -0.5

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
X/C X/C
0.5 18.0% Span 0.5 61.2% Span

Figure 9: Comparison of Shark52 Wing on Stretched Fuselage and Baseline Con guration.

MACH = .780
0.10 0.10
47.5% Span 89.9% Span
SYMBOL SOURCE ALPHA CL CD
SYN88 - Shark52 .718 .2714 .00738
0.05 0.05
SYN88 - Shark1 1.000 .2678 .01796
Y/C

Y/C

0.00 0.00

-0.05 -0.05

Shark52
-0.10 -0.10
Upper-Surface Isobars
0.10 ( Contours at .05 Cp ) 0.10
32.4% Span 75.4% Span

0.05 0.05
Y/C

Y/C

0.00 0.00

-0.05 -0.05

-0.10 -0.10
0.10 0.10
18.0% Span 61.2% Span

0.05 0.05
Y/C

Y/C

0.00 0.00

-0.05 -0.05

-0.10 -0.10

Figure 10: Comparison of Shark52 and Shark1 Wing Drag Loops.

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 20 of 26

 MACH = .780
( Contours at .05 Cp )
SYN88 - Shark52 SYN88 - Shark1
ALPHA = .72 , CL = .2714 ALPHA = 1.00 , CL = .2678
CD = .00738 CD = .01796

Figure 11: Comparison of Shark52 and Shark1 Wing Pressure Contours.

REN = 14.50 , MACH = .780

-1.5 -1.5
SYMBOL SOURCE ALPHA CL CD
SYN107P - SharkNS7 .633 .2605 .01283
-1.0 -1.0
SYN107P - SharkNS7 -.021 .1808 .01128
SYN107P - SharkNS7 1.268 .3404 .01517
Cp

Cp

-0.5 -0.5

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
X/C SharkNS7 X/C
0.5 47.5% Span Upper-Surface Isobars 0.5 89.8% Span
-1.5 ( Contours at .05 Cp ) -1.5

-1.0 -1.0
Cp

Cp

-0.5 -0.5

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
X/C X/C
0.5 32.4% Span 0.5 75.4% Span
-1.5 -1.5

-1.0 -1.0
Cp

Cp

-0.5 -0.5

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
X/C X/C
0.5 18.0% Span 0.5 61.2% Span

Figure 12: Result of Navier-Stokes Inverse Design.

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 21 of 26

 Final Shark Wing
Airfoil Geometry -- Camber & Thickness Distributions
8.0
SYMBOL AIRFOIL ETA R-LE Tavg Tmax @ X
Outboard 68.0 1.055 3.96 5.74 38.62
7.0 Root 18.1 1.488 4.34 6.48 31.51

6.0

5.0
Half-Thickness

4.0

3.0

2.0

1.0

0.0

1.0
Camber

0.0

-1.0

-2.0
10.0
Airfoil

0.0

-10.0
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
Percent Chord

Figure 13: Final Wing Airfoil Geometry - Thickness & Camber Plots.

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 22 of 26

 -1.5 REN = 73.50 , MACH = 0.850 -1.5

-1.0 -1.0
SYMBOL SOURCE ALPHA CL CD
SYN107P BASELINE 0.035 0.450 0.01628
-0.5 SYN107P BASELINE -0.264 0.400 0.01415 -0.5
Cp

Cp
SYN107P BASELINE 0.322 0.500 0.01888
0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
0.5 X/C 0.5 X/C
40.6% Span 90.0% Span

1.0 1.0
BAE WING
-1.5 Upper-Surface Isobars -1.5
( Contours at 0.05 Cp )
-1.0 -1.0

-0.5 -0.5
Cp

Cp
0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
0.5 X/C 0.5 X/C
25.7% Span 74.8% Span

1.0 1.0
-1.5 -1.5

-1.0 -1.0

-0.5 -0.5
Cp

Cp
0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
0.5 X/C 0.5 X/C
9.3% Span 58.2% Span

1.0 1.0

Figure 14: Baseline MDO Datum Wing Pressures at Three Lifting Conditions.

-1.5 REN = 73.50 , CL = 0.450 -1.5

-1.0 -1.0
SYMBOL SOURCE MACH ALPHA CD
SYN107P BASELINE 0.850 0.035 0.01628
-0.5 SYN107P BASELINE 0.840 0.173 0.01506 -0.5
Cp

Cp

SYN107P BASELINE 0.860 -0.133 0.01831

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
0.5 X/C 0.5 X/C
40.6% Span 90.0% Span

1.0 1.0
BAE WING
-1.5 Upper-Surface Isobars -1.5
( Contours at 0.05 Cp )
-1.0 -1.0

-0.5 -0.5
Cp

Cp

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
0.5 X/C 0.5 X/C
25.7% Span 74.8% Span

1.0 1.0
-1.5 -1.5

-1.0 -1.0

-0.5 -0.5
Cp

Cp

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
0.5 X/C 0.5 X/C
9.3% Span 58.2% Span

1.0 1.0

Figure 15: Baseline MDO Datum Wing Pressures at Three Mach Numbers.

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 23 of 26

 -1.5 REN = 73.50 , MACH = 0.850 -1.5

-1.0 -1.0
SYMBOL SOURCE ALPHA CL CD
SYN107P REDESIGN 0.425 0.450 0.01568
-0.5 SYN107P REDESIGN 0.139 0.400 0.01370 -0.5
Cp

Cp
SYN107P REDESIGN 0.694 0.500 0.01834
0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
0.5 X/C 0.5 X/C
40.6% Span 90.0% Span

1.0 1.0
BAE REDESIGN
-1.5 Upper-Surface Isobars -1.5
( Contours at 0.05 Cp )
-1.0 -1.0

-0.5 -0.5
Cp

Cp
0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
0.5 X/C 0.5 X/C
25.7% Span 74.8% Span

1.0 1.0
-1.5 -1.5

-1.0 -1.0

-0.5 -0.5
Cp

Cp
0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
0.5 X/C 0.5 X/C
9.3% Span 58.2% Span

1.0 1.0

Figure 16: Redesigned MDO Datum Wing Pressures at Three Lift Coecients.

-1.5 REN = 73.50 , CL = 0.450 -1.5

-1.0 -1.0
SYMBOL SOURCE MACH ALPHA CD
SYN107P REDESIGN 0.850 0.425 0.01568
-0.5 SYN107P REDESIGN 0.840 0.558 0.01524 -0.5
Cp

Cp

SYN107P REDESIGN 0.860 0.260 0.01665

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
0.5 X/C 0.5 X/C
40.6% Span 90.0% Span

1.0 1.0
BAE REDESIGN
-1.5 Upper-Surface Isobars -1.5
( Contours at 0.05 Cp )
-1.0 -1.0

-0.5 -0.5
Cp

Cp

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
0.5 X/C 0.5 X/C
25.7% Span 74.8% Span

1.0 1.0
-1.5 -1.5

-1.0 -1.0

-0.5 -0.5
Cp

Cp

0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
0.5 X/C 0.5 X/C
9.3% Span 58.2% Span

1.0 1.0

Figure 17: Redesigned MDO Datum Wing Pressures at Three Mach Numbers.

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 24 of 26

 -1.5 REN = 73.50 , MACH = 0.850 , CL = 0.450 -1.5

-1.0 -1.0
SYMBOL SOURCE ALPHA CD
SYN107P REDESIGN 0.425 0.01568
-0.5 SYN107P BASELINE 0.035 0.01628 -0.5
Cp

Cp
0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
0.5 X/C 0.5 X/C
40.6% Span 90.0% Span

1.0 1.0
BAE REDESIGN
-1.5 Upper-Surface Isobars -1.5
( Contours at 0.05 Cp )
-1.0 -1.0

-0.5 -0.5
Cp

Cp
0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
0.5 X/C 0.5 X/C
25.7% Span 74.8% Span

1.0 1.0
-1.5 -1.5

-1.0 -1.0

-0.5 -0.5
Cp

Cp
0.0 0.0
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
0.5 X/C 0.5 X/C
9.3% Span 58.2% Span

1.0 1.0

Figure 18: Comparison of Redesigned and Baseline MDO Datum Wing Pressures.

REN = 73.50 , MACH = 0.850 , CL = 0.450

0.7 SYMBOL SOURCE ALPHA CD

SYN107P REDESIGN 0.425 0.01568
SYN107P BASELINE 0.035 0.01628

0.6

0.5
SPANLOAD & SECT CL

0.4 CL

0.3
C*CL/CREF

0.2

0.1

0.0

-0.1
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0 110.0

PERCENT SEMISPAN

Figure 19: Comparison of Redesigned and Baseline MDO Datum Wing Spanloads.

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 25 of 26

 Airfoil Geometry -- Camber & Thickness Distributions
7.0
SYMBOL AIRFOIL ETA R-LE Tavg Tmax @X
BASELINE 36.8 2.288 3.42 5.00 37.17
REDESIGN 36.8 2.289 3.48 4.98 34.31
6.0

5.0
Half-Thickness

4.0

3.0

2.0

1.0

0.0

2.0
Camber

1.0

0.0

-1.0
10.0
Airfoil

0.0

-10.0
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
Percent Chord

Figure 20: MDO Datum Wing Airfoil Geometry - Thickness & Camber Plots.

Jameson & Vassberg, AIAA Paper 2001-0538, Reno, NV 26 of 26

View publication stats